Anthropic has launched Claude Fable 5, its first publicly available model in the Mythos class, which features a safeguard system that automatically switches to a safer model in three high-risk areas. The Mythos class represents Anthropic’s highest model tier, sharing a base model with the recently unlocked Claude Mythos 5. Fable serves as a mid-tier model with restricted access to safety-critical functions. Previously, access to Mythos systems was limited to over 40 elite technology companies due to misuse risks, resulting from unauthorized access incidents that intensified security discussions.
Claude Fable 5 replaces the earlier model, Claude Opus 4.8, becoming the most capable publicly accessible model. The pricing for Fable 5 is set at $10 per million input tokens and $50 per million output tokens, marking a twofold increase compared to Opus 4.8. In comparison to the Mythos Preview, its cost is less than half of the previous offering, broadening access for more developers. From June 22, 2026, Fable 5 will be available free for users on Pro, Max, Team, and Enterprise plans, with immediate access via API.
Fable 5 has been generally accessible in GitHub Copilot since June 9, 2026, facilitating swift integration into development environments. The model incorporates a safeguard architecture that routes requests in high-risk areas—cybersecurity, biology and chemistry, and model capability extraction—to the older Claude Opus 4.8. According to Anthropic, this fallback mechanism engages in fewer than 5% of sessions. Tests did not reveal any universal jailbreaks after over 1,000 hours of external red teaming.
All Mythos-class traffic is subject to a 30-day data retention rule and will not contribute to model training. Anthropic positions itself as the first provider to implement a formal structure for the controlled release of models deemed too dangerous for unrestricted public access. Claude Mythos 5, with some safeguards lifted, is available to selected partners such as AWS, Microsoft, Apple, and CrowdStrike, as well as approved bio-researchers.
In benchmark comparisons, Fable 5 exhibits substantial improvements over Opus 4.8. It achieves 95.0% on software engineering benchmarks like SWE-bench Verified, compared to Opus’s 88.6%. On SWE-bench Pro, Fable 5’s lead increases to 80.3% versus 69.2%. Specialized tests reveal Fable 5 scored 29.3% on FrontierCode Diamond, compared to 13.4% for Opus 4.8. In practical applications, Stripe migrated over 50 million lines of Ruby code in one day utilizing Fable 5’s capabilities.
Additionally, the ExploitBench results indicate a significant performance gap in offensive capabilities, with Mythos 5 scoring 78.0%, in contrast to 40.0% for Opus 4.8. This performance gap highlights the importance of the fallback system integrated within Fable 5’s public access model. On June 1, 2026, Anthropic confidentially filed a draft S-1 form with the SEC for an IPO, valuing the company at $965 billion after a $65 billion funding round. The revenue run rate was approximately $47 billion as of May 2026, anticipated to exceed $50 billion by the end of June, boosted by significant growth from a run rate of $4 billion in July 2025. Coinciding with this filing, OpenAI also confidentially submitted for a listing on June 8, 2026, indicating a potential maturation phase in the AI sector.
